Mengdi Yang
Hot-wire Assisted
Atomic Layer Deposition
of Tungsten Films
Mengdi Yang
m.yang@utwente.nl
Invitation
You are cordially
invited to the public
defense of my PhD
thesis entitled:
Hot-wire Assisted
Atomic Layer Deposition
of Tungsten Films
Friday
February 2
nd, 2018
16:30 Introduction
16:45 Public defense
Prof. Dr. G. Berkhoffzaal,
in building Waaier,
University of Twente
Paranymphs:
Xingyu Liu
Ramazan Oguzhan
Apaydin
Hot-wire
Assisted
Atomic Layer Deposi
tion of T
ungsten Films
Mengd
i Y
ang
ISBN: 978-90-365-4469-6
HOT-WIRE ASSISTED
ATOMIC LAYER
DEPOSITION OF
TUNGSTEN FILMS
Mengdi Yang
UNIVERSITY OF TWENTE
Ph.D. Graduation Committee:
Chairman: prof. dr. P.M.G. Apers University of Twente
Secretary: prof. dr. P.M.G. Apers University of Twente
Supervisor: prof. dr. J. Schmitz University of Twente
Co-supervisor: dr. A.K. Kovalgin University of Twente
Committee members: prof. dr. ir. J.R. van Ommen Delft University of
Technology
dr. J.-W. Maes ASM International N.V.
prof. dr. ir. R.A.M. Wolters University of Twente
prof. dr. F. Roozeboom Eindhoven University of
Technology
prof. dr. ing. A.J.H.M. Rijnders University of Twente
This research was funded by NWO Domain Applied and Engineering Sciences (NWO-TTW), project “plasma-free atomic layer deposition”, nr. 12846 and carried out at MESA+ Institute for Nanotechnology, University of Twente.
Title: Hot-wire assisted atomic layer deposition of tungsten films ISBN: 978-90-365-4469-6
DOI: 10.3990/1.9789036544696
https://doi.org/10.3990/1.9789036544696
Copyright © 2018 by Mengdi Yang, Enschede, The Netherlands. All rights reserved
Front cover: TEM images of a 13 nm W film grown by HWALD inside Al2O3-coated Si
trenches (Chapter 4).
Back cover: Totoro, source: https://www.drawingtutorials101.com/how-to-draw-totoro-from-my-neighbor-totoro
HOT-WIRE ASSISTED ATOMIC LAYER
DEPOSITION OF TUNGSTEN FILMS
DISSERTATION
to obtain
the degree of doctor at the University of Twente, on the authority of the rector magnificus,
prof. dr. T.T.M. Palstra,
on account of the decision of the graduation committee, to be publicly defended
on Friday, 2nd February, 2018 at 16.45
by
Mengdi Yang
born on 10th February, 1989
This dissertation is approved by
prof. dr. J. Schmitz (promoter)
To my family
La vie, voyez-vous, ça n'est jamais si bon ni si mauvais qu'on croit. ---- Guy de Maupassant 《Une Vie》 生活不可能像你想象得那么好,但也不会像你想象得那么糟
——莫泊桑《人生》(Translated into Chinese)
人的脆弱和坚强都超乎自己的想象。有时,可能脆弱得一句话就泪流满面,有时, 也发现自己咬着牙走了很长的路。
Contents
1 Introduction ... 1
1.1. Motivation of the project ... 2
1.2. Introduction to atomic layer deposition ... 3
1.3. From thermal to radical enhanced ALD ... 3
1.4. Application of HWALD in tungsten deposition ... 5
1.5. Outline ... 6
2 HWALD reactors: design and supply of atomic hydrogen ... 11
2.1. Thin film deposition facilities ... 12
2.1.1. Cold-wall reactor ... 13
2.1.2. Hot-wall reactor ... 14
2.2. Hot-wire temperature calibration ... 15
2.3. Spectroscopic ellipsometry ... 16
2.3.1. Optical model used for SE ... 17
2.3.2. Thickness verification ... 18
2.4. Delivery of at-H ... 20
2.4.1. Te etching by at-H pulses in the cold-wall reactor ... 21
2.4.2. Factors influencing on Te etch rate ... 23
2.4.3. Back-stream diffusion ... 24
2.5. Hot wire is not a source for W deposition ... 26
2.6. Conclusions ... 27
3 High resistivity β-phase HWALD W grown in a cold-wall reactor ... 29
3.1. Introduction ... 30
3.2. Experimental ... 30
3.2.1. Deposition ... 30
3.2.2. Film characterization ... 31
3.3. In-situ study of the interplay between CVD, etching and ALD modes ... 32
3.3.2. Etching mode of deposited W films ... 36
3.3.3. Optimization towards the dominant HWALD process ... 39
3.3.4. Successful HWALD W ... 40
3.3.5. Conclusion ... 43
3.4. A Comparison of Tungsten Films Grown by CVD and HWALD ... 43
3.4.1. Deposition of tungsten by three methods ... 43
3.4.2. Resistivity and density ... 44
3.4.3. Surface roughness ... 45
3.4.4. Crystallinity ... 49
3.4.5. Conclusions ... 53
3.5. Wrap up ... 54
4 Low-resistivity α-phase HWALD W grown in a hot-wall reactor ... 57
4.1. Introduction ... 58
4.2. Experimental ... 58
4.2.1. Deposition of HWALD tungsten films ... 58
4.2.2. Ex-situ analysis ... 59
4.2.3. Resistivity measurements ... 59
4.2.4. Film continuity... 60
4.3. Results and discussion ... 61
4.3.1. HWALD window ... 61
4.3.2. Surface roughness and film composition ... 65
4.3.3. Crystallinity ... 66
4.3.4. Resistivity and uniformity ... 70
4.3.5. Step coverage on HAR substrates ... 73
4.3.6. Continuity ... 76
4.4. Conclusions ... 77
5 Effects of oxygen, nitrogen and fluorine on the formation of α- and β-phase W by HWALD ... 81
5.2. Methodology to study the influence of additional gases ... 82
5.3. Results and discussion ... 83
5.3.1. Influence of N2O and O2 on the HWALD W process ... 83
5.3.2. Influence of H2O vapor on the HWALD W process ... 87
5.3.3. Growth recovery after oxidation and its termination by nitrogen-containing gases ... 88
5.3.4. Influence of WF6 overdose on the HWALD W process ... 91
5.4. Conclusions ... 93
6 Inherently area-selective HWALD of W films on metal/insulator substrates ... 95
6.1. Introduction ... 96
6.2. Experimental ... 97
6.3. Results and discussion ... 99
6.3.1. Nucleation of HWALD W on substrates of various materials ... 99
6.3.2. Selective growth of HWALD W ... 104
6.3.3. Nucleation of HWALD W on a-Si seed layer of various thicknesses ... 108
6.4. Conclusions ... 110 7 Conclusions ... 113 7.1. Summary ... 114 7.2. Outlook ... 117 List of Publications ... 119 Acknowledgement ... 123
1
2
1.1. Motivation of the project
Currently, shrinking dimensions of modern ultra-large-scale integration (ULSI) pose stringent demands on the control of film conformity, uniformity and thickness. Moreover, the increasingly complex structures require a uniform step coverage when fabricating nano-scale features in high-aspect-ratio and even three-dimensional (3D) structures. In this light, conventional physical vapour deposition (PVD) and chemical vapour deposition (CVD) have their limitations because of their
poor step coverage and insufficient control of thickness[1,2].
In this project, metallic layers (films) are our focused materials because they play a crucial role as electrodes and interconnects in integrated circuits such as microprocessors, dynamic random-access memories, flash memories, and image sensors. Commonly applied metals include aluminum (Al), copper (Cu), titanium (Ti) and tungsten (W). Pure-metallic films are traditionally deposited by PVD, i.e.
sputtering or evaporation, or by CVD[2-4]. However, as mentioned above, memories
and logic ULSIs having a three-dimensional structure are shrinking. For example, the half pitch of the Cu interconnects used in ULSIs, fabricated by
through-silicon-via technology, are expected to be reduced to 15 nm[5]. Fin Field Effect Transistors
are commonly selected for high performance and low power consumption circuits[6].
These technologies require a novel technique of metal deposition to achieve downscaled metal layers of high-aspect-ratios and even 3D features.
To this purpose, atomic layer deposition (ALD) has emerged as an important technique for depositing ultra-thin metal films. The most prominent advantage of ALD over other deposition techniques is its self-limiting reaction mechanism, the main feature of ALD, leading to an accurate thickness control and high film uniformity on
arbitrarily shaped and patterned surfaces[7]. The combination of these two features
is demanded for e.g. future 3D integration. A reduced processing temperature, natural to ALD, additionally allows the manufacturing of devices by post-processing of electronic chips.
3
1.2. Introduction to atomic layer deposition
ALD is a technique suitable for manufacturing inorganic material layers with thickness down to a fraction of one monolayer. It is based on a sequential use of
self-limiting gas-solid reactions[8]. During deposition, two reactive gases A and B,
also known as precursors, are introduced sequentially to the reactor in order to react at the substrate surface. The duration of gas introduction is normally short in the form of a pulse. Different from CVD, a reactor purge in between precursor pulses avoids the mixture of the reactants in the gas phase. Thus a standard ALD cycle can be expressed as (a) precursor A/ (b) purge/ (c) precursor B/ (d) purge. Normally, in (a) and (c) the two precursors must have self-limiting reactions on the surface, meaning the precursor can neither react nor adsorb on a surface where this surface is already covered by this precursor. In this manner, saturation will occur when introducing one precursor superfluously. Furthermore, purges in (b) and (d) need to be long enough to ensure a sufficient removal of residual precursors. As a result, each reaction cycle adds a given amount of material to the surface, referred to as the growth per cycle (GPC). To grow a desired layer thickness, the ALD cycle is repeated.
An ALD window is defined as the region where an ALD process fulfills the
criterion of self-limiting reactions[8]. This window can be achieved in a certain range
of experimental conditions, within which the GPC is constant (for example a substrate-temperature window). With this constant GPC, a film will have a linear growth by repeated ALD cycles. Hence, finding the window is crucial in studies of ALD.
1.3. From thermal to radical enhanced ALD
ALD was conventionally developed to deposit two-element materials such as oxides or nitrides. Single-element films of metals and semiconductors, with a few
exceptions, are however difficult to deposit using thermal-only ALD processes[7]. A
solution in this case is called radical-enhanced ALD (REALD)[1], which normally
4
species (radicals). REALD is used to enable deposition of films at much lower
substrate temperatures (Ts) than thermal ALD. For example, radical-enhanced
aluminum oxide (Al2O3) ALD can be performed using TMA and Oradicals at a Ts as
low as room temperature[9]. The most common REALD variety is plasma-enhanced
ALD (PEALD). Hydrogen- or nitrogen-based plasmas have been used to deposit
tantalum (Ta), titanium (Ti), ruthenium (Ru), silicon (Si) and germanium (Ge)[7], Al[10]
as well as aluminum nitride (AlN)[11] and gallium nitride (GaN)[12]. With PEALD,
platinum (Pt) can be grown by oxygen (O2) plasma[13], Ru[14,15] and cobalt (Co) by
ammonia (NH3) plasma[16], and Al[10] and Cu[17] by hydrogen (H2) plasma.
However, some limitations make PEALD less attractive. The first is a reduced step coverage compared to thermal ALD, caused by radical recombination away from the plasma. For example, the recombination of hydrogen radicals on the walls of trenches attenuates the hydrogen radical flux in the downward direction, resulting
in a non-uniform Ta film formed in trenches with an aspect ratio of 40:1[18,19]. The
second limitation is that plasmas cause damage to the wafer under treatment, in particular to metal-oxide-semiconductor (MOS) transistors. A common solution is to employ a remote plasma, by increasing distance between the plasma source and the substrate. This, however, will increase the reactor volume leading to long ALD cycle times. As the distance increases, radical recombination can further occur during their delivery to the wafer surface. Additionally, a plethora of radicals is typically created in even simple one- or two-gas plasmas, enabling numerous chemical reactions besides the desired one. Not all the reactions lead to the formation of high-quality films; certain plasma compounds can deteriorate film quality. As a result, the wafer surface is exposed to various ions, radicals and atoms, as well
as ultra-violet (UV) photons[20]. This makes the composition and structure of the
growing film hard to predict and control.
Apart from plasma, there are several other methods to generate radicals.
There are reports about radicals created by means of a supersonic jet[21-23]. Moreover,
some molecules, such as methane (CH4), NH3 and H2, have been found to dissociate
on surfaces of transition metals[24-27]. More promisingly, the hot-wire CVD (HWCVD)
has been well established using a heated metal filament, also called hot wire (HW),
5
deposit various materials including poly-Si[30], Si
3N4[31] and metal oxides[32,33].
Enlightened by HWCVD, in this work, we propose to utilize hot wire for generating radicals for REALD processes, instead of using a plasma. In view of catalytic activities of metals, there are many potential candidates to realize hot wires,
such as Pt[34], nickel (Ni)[35] and W[36]. In this study, tungsten (W) filament became
our choice due to its long durability and low cost.
It is well-established that radicals can form upon catalytic dissociation of
certain gas molecules on a hot W filament[37-40]. For instance, molecular hydrogen
(H2)[41-43] and NH3[44] can effectively decompose on a W filament heated up to 1500-2000 °C. By replacing plasma by a hot wire, we aim to avoid substrate damage, limit the number of radicals formed in gas phase and thereby achieve a better control of the reactant supply. The hot-wire assisted ALD (HWALD) is a novel technique explored in only a few publications so far. The available works on HWALD concern
Co and Co(W) films[45,46], where NH
3 dissociated upon a hot wire is used as a
reducing agent to form these metals. Kostis et al.[47] introduced oxygen gas to oxidize
hot tungsten filament and form volatile tungsten oxides (WOx). The oxides were
further transported to the substrate, resulting in WOx deposition. Ni was reported to
be deposited by cracking NH3 on a hot wire[48,49]. In this work, we aim to propose and
establish a HWALD process for deposition of tungsten films.
1.4. Application of HWALD in tungsten deposition
In today’s metallization schemes, W vias are widely used to provide inter-level contacts between metal layers due to tungsten’s inertness to many chemicals,
compatibility with silicon technology and low electrical resistivity[50,51]. Tungsten can
conventionally be deposited by chemical vapor deposition (CVD) or by
sputtering[52,53]. To meet the requirements of recent ULSI, ALD is rapidly
strengthening its position as a method suitable for industrial use[54].
Successful ALD of W has been reported by using tungsten hexafluoride
6
The reductants form an intermediate sacrificial layer (i.e. Si or boron (B),
respectively), which can be turned into W while reacting with WF6. However,
deposition of Si or B in those cases is hardly limited to one-monolayer formation, thereby diminishing the self-limiting nature of ALD. It may additionally leave Si, B
and fluorine (F) impurities inside the films, resulting in a higher-resistivity W[61].
In this project, we focus on the deposition of tungsten films by HWALD using
sequential pulses of WF6 gas and atomic hydrogen (at-H). The latter is generated by
dissociation of H2 on a hot wire, as described in the previous section. It is known
from the literature that the dissociation probability of H2 rises with increasing hot-wire
temperature, and can reach a maximum at around 2700 °C[62]. Importantly, the hot
wire (HW) itself is not a source of tungsten: the W vapor pressure is rather low at
temperatures below 2000 °C[63]. Therefore the filament is not an efficient source of
W. This will be further discussed in section 2.5 of Chapter 2. In the HWALD of W, at-H acts as a reductant to remove fluorine (F) termination of the growing film by forming HF gas, leaving pure W surface. This work explores HWALD W in view of its potential use in backend-of-line processing of future ULSI circuits.
1.5. Outline
In this thesis, we demonstrate the novel HWALD technique and successful realization of tungsten (W) films of supreme properties by this technique. In Chapter 2, two deposition reactors are described. The spectroscopic ellipsometry (SE) used to in-situ monitor the film growth and the optical models for grown layers are explained. In addition, the delivery of at-H to the substrate is confirmed by tellurium (Te) etching experiments. In Chapter 3, results of HWALD W formed in the cold-wall reactor are presented. In that reactor, apart from HWALD, the co-existent parasitic CVD and etching of the grown films were also found and the interplay among them depended on experimental conditions. After optimizing the HWALD process by suppressing CVD and etching, high-purity W films were successfully formed.
However, the deposited W films were grown in β-phase, possessing a higher
7
In order to obtain low-resistivity W, we started to utilize the hot-wall reactor.
The results are demonstrated in Chapter 4. W grown in α-phase has been
successfully obtained in the hot-wall reactor. The resistivity of this W is as low as
15 µΩ·cm. Moreover, the HWALD W can be deposited in trenches with an aspect
ratio as high as 36. To investigate factors leading to the preferential formation of
either α- or β-phase, some additional reactive gases were introduced during the
HWALD process, as described in Chapter 5. It has been found that oxidants such
as N2O, O2 and water (H2O) vapor have profound effects on GPC, probably due to
surface oxidation. However, the oxidized W could still be reduced by at-H, continuing
the HWALD process and leaving the HWALD W of high purity and in the α-phase
form. In opposite, nitrogen-containing gases, i.e., N2O and NH3, effectively
terminated the growth of HWALD W. Apart from the oxidants and nitrogen-containing gases, the surplus of fluorine-containing gas appeared to be a crucial factor leading
to the formation of β-phase W instead of α-phase.
In Chapter 6, we characterize and compare the nucleation and growth of HWALD W films on various substrates. As a result, an inherently area-selective
HWALD W process was achieved on W/SiO2 and Co/SiO2 patterned surfaces.
Furthermore, ultra-thin a-Si seed layers were explored in order to start HWALD of W on surfaces showing inhibited nucleation. Applying an a-Si seed layer far below 1 nm in thickness appeared to be sufficient to support the effective nucleation, enabling standard GPC with little to no incubation time.
References
[1] H. Kim, J. Vac. Sci. Technol. B 21, 2231 (2003).
[2] T. T. Kodas and M. J. Hampden-Smith, The chemistry of metal CVD
(Wiley-VCH Verlag GmbH, Weinheim, 1994).
[3] J. D. Plummer, M. D. Deal and P. B. Griffin, Silicon VLSI technology:
fundamentals, practice, and modeling (Prentice Hall, Upper Saddle River, 2000).
[4] W. Kern and J. L. Vossen, Thin film processes II, Vol. 2 (Academic Press,
San Diego, 2012).
[5] H. Shimizu, K. Sakoda, T. Momose, M. Koshi and Y. Shimogaki, J. Vac.
8
[6] The International Technology Roadmap for Semiconductor, 2015 Update
http://www.itrs2.net/itrs-reports.html.
[7] S. M. George, Chem. Rev. 110, 111 (2010).
[8] R. L. Puurunen, J. Appl. Phys. 97, 9 (2005).
[9] S. Heil, P. Kudlacek, E. Langereis, R. Engeln, M. van de Sanden and W.
Kessels, Appl. Phys. Lett. 89, 131505 (2006).
[10] Y. J. Lee and S. W. Kang, Electrochem. Solid St. Lett. 5, C91 (2002).
[11] Y. J. Lee and S. W. Kang, Thin Solid Films 446, 227 (2004).
[12] C. Ozgit, I. Donmez, M. Alevli and N. Biyikli, J. Vac. Sci. Technol. A 30,
01A124 (2012).
[13] D. Longrie, K. Devloo-Casier, D. Deduytsche, S. van den Berghe, K.
Driesen and C. Detavernier, J. Solid State Sci. Technol. 1, Q123 (2012).
[14] O.-K. Kwon, S.-H. Kwon, H.-S. Park and S.-W. Kang, J. Electrochem. Soc.
151, C753 (2004).
[15] O.-K. Kwon, S.-H. Kwon, H.-S. Park and S.-W. Kang, Electrochem. Solid
St. Lett. 7, C46 (2004).
[16] H. Kim, Electrochem. Solid St. Lett. 9, G323 (2006).
[17] J. Mao, E. Eisenbraun, V. Omarjee, A. Korolev, C. Lansalot and C.
Dussarrat, ECS Trans. 33, 125 (2010).
[18] R. Grubbs and S. George, J. Vac. Sci. Technol. A 24, 486 (2006).
[19] H. Kim, C. Cabral Jr, C. Lavoie and S. Rossnagel, J. Vac. Sci. Technol. B
20, 1321 (2002).
[20] V. V. Afanas`ev, J. M. M. de Nijs and P. Balk, J. Appl. Phys. 78, 6481
(1995).
[21] P. Misra, X. Zhu, C.-Y. Hsueh and J. B. Halpern, Chem. Phys. 178, 377
(1993).
[22] H. Fu, Y. Hu and E. Bernstein, J. Chem. Phys. 125, 014310 (2006).
[23] D. W. Kohn, H. Clauberg and P. Chen, Rev. Sci. Instrum. 63, 4003 (1992).
[24] G. Papapolymerou and V. Bontozoglou, J. Mol. Catal. A: Chem. 120, 165
(1997).
[25] J. Harris and S. Andersson, Phys. Rev. Lett. 55, 1583 (1985).
[26] J. Harris, J. Simon, A. Luntz, C. Mullins and C. Rettner, Phys. Rev. Lett.
67, 652 (1991).
[27] M.-S. Liao and Q.-E. Zhang, J. Mol. Catal. A: Chem. 136, 185 (1998).
[28] H. Matsumura, H. Umemoto and A. Masuda, J. Non-Cryst. Solids 338, 19
(2004).
[29] R. Schropp, Thin Solid Films 395, 17 (2001).
[30] J. Rath, H. Meiling and R. Schropp, Sol. Energy Mater. Sol. Cells 48, 269
(1997).
[31] P. Alpuim, L. Gonçalves, E. S. Marins, T. Viseu, S. Ferdov and J. Bourée,
Thin Solid Films 517, 3503 (2009).
[32] S. Mitra, K. Sridharan, J. Unnam and K. Ghosh, Thin Solid Films 516, 798
(2008).
[33] C. M. White, D. T. Gillaspie, E. Whitney, S.-H. Lee and A. C. Dillon, Thin
Solid Films 517, 3596 (2009).
9
[35] J. Thangala, S. Vaddiraju, R. Bogale, R. Thurman, T. Powers, B. Deb and
M. K. Sunkara, Small 3, 890 (2007).
[36] J. K. Holt, M. Swiatek, D. G. Goodwin and H. A. Atwater, J. Appl. Phys. 92,
4803 (2002).
[37] A. H. Mahan, A. C. Dillon, L. M. Gedvilas, D. L. Williamson and J. D.
Perkins, J. Appl. Phys. 94, 2360 (2003).
[38] A. H. Mahan, J. Carapella, B. P. Nelson, R. S. Crandall and I. Balberg, J.
Appl. Phys. 69, 6728 (1991).
[39] R. Toukabri, N. Alkadhi and Y. Shi, J. Phys. Chem. A 117, 7697 (2013).
[40] Y. Shi, Acc. Chem. Res. 48, 163 (2015).
[41] I. Langmuir, J. Am. Chem. Soc. 34, 860 (1912).
[42] I. Langmuir and G. M. J. Mackay, J. Am. Chem. Soc. 36, 1708 (1914).
[43] I. Langmuir, J. Am. Chem. Soc. 37, 417 (1915).
[44] H. Matsumura and H. Tachibana, Appl. Phys. Lett. 47, 833 (1985).
[45] H. Shimizu, K. Sakoda, T. Momose and Y. Shimogaki, Japanese Journal of
Applied Physics 51, 05EB02 (2012).
[46] H. Shimizu, K. Sakoda, T. Momose, M. Koshi and Y. Shimogaki, Journal of
Vacuum Science & Technology A 30, 01A144 (2012).
[47] I. Kostis, M. Vasilopoulou, G. Papadimitropoulos, N. Stathopoulos, S.
Savaidis and D. Davazoglou, Surface and Coatings Technology 230, 51 (2013).
[48] G. Yuan, H. Shimizu, T. Momose and Y. Shimogaki, J. Vac. Sci. Technol.
A 32, 01A104 (2014).
[49] G. Yuan, H. Shimizu, T. Momose and Y. Shimogaki, Microelectron. Eng.
120, 230 (2014).
[50] E. Lassner and W.-D. Schubert, Tungsten: Properties, chemistry,
thechnology of the element, alloys, and chemical compounds (Kluwer Academic/ Plenum Publishers, New York, 1999).
[51] E. Lassner and W. D. Schubert, Tungsten: Properties, Chemistry,
Technology of the Elements, Alloys, and Chemical Compounds (Springer Science & Business Media, New York, 1999).
[52] C. H. Ting and M. Paunovic, J. Electrochem. Soc. 136, 456 (1989).
[53] S. Rossnagel, I. Noyan and C. Cabral Jr, J. Vac. Sci. Technol. B 20, 2047
(2002).
[54] M. Leskelä and M. Ritala, Thin Solid Films 409, 138 (2002).
[55] J. Klaus, S. Ferro and S. George, Thin Solid Films 360, 145 (2000).
[56] J. Elam, C. Nelson, R. Grubbs and S. George, Surf. Sci. 479, 121 (2001).
[57] J. Elam, C. Nelson, R. Grubbs and S. George, Thin Solid Films 386, 41
(2001).
[58] S. A. Van Norman, J. W. Tringe, J. D. Sain, R. Yang, J. L. Falconer and A.
W. Weimer, Appl. Phys. Lett. 106, 153102 (2015).
[59] B. Kalanyan, M. D. Losego, C. J. Oldham and G. N. Parsons, Chem. Vap.
Deposition 19, 161 (2013).
[60] G. Wang, Q. Xu, T. Yang, J. Xiang, J. Xu, J. Gao, C. Li, J. Li, J. Yan and D.
Chen, ECS J. Solid State Sci. Technol. 3, P82 (2014).
[61] S.-H. Kim, N. Kwak, J. Kim and H. Sohn, J. Electrochem. Soc. 153, G887
10
[62] J. N. Smith Jr. and W. L. Fite, J. Chem. Phys 37, 898 (1962).
[63] R. Szwarc, E. Plante and J. Diamond, NBS J. Res. Phys. Chem. 69, 417
11
2
Abstract.
This chapter descibes the hot- and cold-wall reactors used in the experiments to deposit HWALD W films. Basic characteristics of both reactors are given. We also present the optical model for spectroscopic ellipsometry to extract thickness, resistivity and optical properties of W films, along with the thicknesses verification by other techniques. Further, the existence of at-H and its delivery to the substrate in both reactors are demonstrated: we provide results of telllurium (Te) etching experiments by at-H pulses in the cold-wall reactor. Finally, the influence of several experimental parameters on the etch rate of Te is investigated for optimization of the HWALD process.
12
2.1. Thin film deposition facilities
Two differently designed HWALD reactors have been utilized in the experiments: a big-volume cold-wall reactor and a small-volume hot-wall reactor. The reactors were connected to the home-built cluster system, sharing the gas
network[1] and the loadlock with a base pressure of 10-7 mbar. The loadlock enabled
wafer transfer without vacuum break, ensuring no interface deterioration. The reactors were equipped with a spectroscopic ellipsometer (SE), Woollam M-2000 of
J. A. Woollam Co., Inc., combined with CompleteEASE® software. A hot wire,
implementing resistive heating of a pure tungsten (W) filament, was installed using the special assembly. To bear in mind (see Section 1.4), the hot wire itself was not acting as a source of tungsten.
Figure 2.1. A schematic view of the cluster system. Both reactors used in this thesis are connected to the
13
2.1.1. Cold-wall reactor
A schematic cross-section of the cold-wall HWALD reactor is shown in Fig.
2.2. Molecular hydrogen (H2) carried by argon gas (Ar) was introduced into the
chamber via the hot-wire port (top). The distance between the at-H outlet and the
substrate is approximately 70 cm. WF6 is supplied through the gas ring, situated
around 10 centimeters above the substrate. The volume of this reactor is approximately 70 liters. The advantage of this design, despite a long diffusion distance required, was that the wafer could be protected from the hot-wire radiation.
This reactor could be evacuated to a base pressure of 10-7 mbar by a turbomolecular
pump. The maximum temperature of the substrate holder was 400 °C. High-speed ALD valves were used, with a time resolution of 0.1 s.
Figure 2.2. A schematic cross-sectional view of the cold-wall HWALD reactor. The spectroscopic
14
2.1.2. Hot-wall reactor
The home-built hot-wall reactor[1,2] used for all depositions is schematically
shown in Fig. 2.3. This 24-ml hot-wall inner reactor is placed inside a big (several liters) cold-wall outer reactor. In opposite to the cold-wall version, walls of the inner reactor are heated to reduce gas adsorption on them. The hot wire is installed on the side (see “hot wire” in Fig. 2.3), 2-3 cm away from the wafer. Importantly, there is no direct line-of-sight between the hot wire and the substrate. The second precursor
(WF6) is supplied via the lateral gas inlets distributed around the substrate. This
reactor can be evacuated to a base pressure of 10-7 mbar by a turbomolecular pump.
The maximum temperature of the substrate holder can be as high as 400 °C. Different from its cold-wall counterpart, this hot-wall reactor is not equipped with ALD valves. Instead, there are by-pass lines to evacuate reactive gas when the valve in between the precursor sources and the reactor is closed. Therefore, a constant flow rate of precursors can be maintained during experiments. The opening and closing times of all the valves are programmable with a resolution of 0.1 s.
15
2.2. Hot-wire temperature calibration
In our experiment, the hot-wire radiative temperature was controlled by applying a voltage across it. The hot-wire temperature at around 1500 °C was first measured by a pyrometer assuming the same emissivity of W of the hot wire and W of the pyrometer filament. This point at 1500 °C was further used for the temperature calibration. Fig. 2.4 (a) shows the hot-wire temperature as a function of applied power. The other points besides 1500 °C in this figure were obtained on basis of the known hot-wire geometry and temperature coefficient of resistance (TCR) of
tungsten[3], where the hot-wire resistance was measured at each applied voltage and
current. The film deposition experiments were conducted in the power range of 138 to 336 W, corresponding to the temperature range of 1500 to 1950 °C. Fig. 2.4 (b) presents two I-V curves measured for the hot wire on two different days. The good overlapping verifies a stable performance of this hot wire and indicates the reproducibility of the same hot-wire temperature when applying the same voltage. It is noteworthy that the hot wire exhibited a good stability and reproducibility of the calibration curve after hundreds of hours operation.
(a) (b)
Figure 2.4. (a) A curve presenting the relationship between hot-wire temperature (extracted from the
resistance/TCR values while calibrating one point at 1500 °C) and applied power. (b) Stability of the
16
2.3. Spectroscopic ellipsometry
Spectroscopic ellipsometry (SE) is a non-destructive optical technique that
has been widely used for studying various deposition and etching processes[4-6]. It
measures a change in polarization as light reflects from the wafer surface after passing through the film(s). The polarization change is represented as an amplitude ratio, i.e. Psi (Ψ), and the phase difference, i.e. Delta (Δ). Based on the polarization change, the complex dielectric function of the film(s) can be extracted, providing wavelength-dependent values of the refractive index (n) and the extinction coefficient (k). The measured polarization change depends on the optical properties and the thicknesses of individual materials present on the wafer. Thus, ellipsometry is primarily used to determine film thickness and optical constants. However, it can be also applied to characterize many other properties of measured materials, such as crystallinity, see Fig. 4.7, or resistivity, see Fig. 4.11 in Chapter 4. Moreover, the
substrate temperature (Ts) could be measured by SE using the known temperature
dependence of optical constants of certain materials, such as a Si (100) wafer. Fig.
2.5 compares the readings of Ts by a thermocouple and SE. All temperature values
were recorded after the temperature became stable according to SE. It can be seen from Fig. 2.5 that there was a slight temperature difference of maximum 30 ºC between the two methods.
17
In this thesis, we focus on thickness and electrical properties of W measured by SE. Electrical resistivity can be extracted from the dielectric functions by analyzing
the Drude term[7,8]. The in-situ SE enables thickness monitoring in real time during
deposition, thus providing a close eye to the process of ALD. It appears from both experiments and simulations that a W film thinner than 30 nm is sufficiently transparent for SE measurements.
All in-situ SE measurements shown in this thesis were obtained by using a Woollam M2000 spectroscopic ellipsometer, operating in a wavelength range between 245 and 1688 nm with a resolution of 1.6 nm, in combination with
CompleteEASE® modeling software.The measurement data were acquired
throughout the growth process every 2.5 seconds.
2.3.1. Optical model used for SE
The optical SE model, used to extract thickness and optical constants of W layers, is shown in Table 2.1. In this model, the required optical constants of the Si
substrate, SiO2 and amorphous Si (a-Si) were provided by the software database;
the thicknesses were fitted by SE. The surface roughness was obtained by the
Bruggeman effective medium approximation (EMA)[8], comprising 50% of voids and
50% of the corresponding material.
As the optical constants of thin HWALD W were unknown, its dielectric
functions were parameterized by Drude-Lorentz model[9] with a number of fitting
parameters. To be more specific, the dielectric functions were first constructed, assuming a number of Lorentz oscillators, and then fitted to optical properties obtained by SE. The W seed layers were modelled by introducing 6 Lorentz
oscillators[10]. However, the dielectric functions of HWALD W layers consisted of only
2 Lorentz oscillators and one Drude term[9,11]. The Drude term describes the
intraband absorption of conduction electrons whereas the interband transitions are characterized by Lorentz oscillators. The layer’s conductivity can be extracted from
18
Table 2.1 Layer stack, included in the optical SE model to extract thickness of the grown W.
Layer Material chosen in the model Fitting parameters
Roughness Bruggeman EMA Thickness
Capping layer a-Si[12] Thickness
ALD W layer Parameterized Thickness, parameters of Lorentz oscillators and Drude term
W seed layer Parameterized Thickness, parameters of Lorentz oscillators and Drude term
SiO2 layer SiO2[13] Thickness
Si substrate Si substrate[14] Temperature
the Drude term using the CompleteEase® software.
2.3.2. Thickness verification
To validate the model shown in Table 2.1, for selected samples the thickness of each layer measured by SE was verified by X-ray reflectometry (XRR) and high-resolution scanning electron microscopy (HR-SEM).
Fig. 2.6 shows cross-sectional HR-SEM images of a HWALD tungsten sample with an a-Si capping layer of approximately 5 nm. Thicknesses of the W seed layer and HWALD W layer are roughly 3 nm and 13 nm (by SE), respectively. Fig. 2.6 (a) presents the normal InLens image where the a-Si capping layer cannot be distinguished; the total thickness of all sub-layers (approx. 20 nm by SE) is in agreement with the value of 17 nm obtained by HR-SEM. The energy selective backscattered (ESB) image in Fig. 2.6 (b) exhibits a better contrast between the a-Si and the W layer underneath, giving a total thickness of the two tungsten layers (bright-grey) of approx. 12 nm. Furthermore, Fig. 2.6 (a) indicates a remarkable surface roughness (see further Section 3.4.3). Considering the roughness of a few nm and different measurement positions on the wafer for SE and HR-SEM, the thickness determined by SE is in good agreement with that measured by HR-SEM.
19
(a) (b)
Figure 2.6. Cross-sectional HR-SEM images of a HWALD tungsten film. The tungsten layer is covered
by an approximately 5 nm thick a-Si layer. Image (a) is the standard Inlens detection, showing the total thickness of all tungsten and a-Si sub-layers of approximately 17 nm. Image (b) is the result of Energy Selective Backscattered detection, where approximately 12 nm thick W layers (bright-grey) can be visualized. Pa is the measured thickness (i.e. length of the ↕-line); Pb is the angle between the ↕-line and the substrate surface.
Table 2.2. A comparison of thicknesses measured by SE and XRR at the center of the wafer.
Sample Thickness by in-situ SE
(nm) Thickness by XRR (nm)
1 10.2 9.7
2 24.6 20.2
3 13.9 11.2
Table 2.2 compares the SE and XRR thicknesses of three different samples. The thicknesses measured by XRR are on average 20% smaller than those obtained by SE. To note, the SE data are measured in the center of the wafer, sometimes without having a-Si capping layer on W, whereas the XRR measurements were always done on the wafer with the capping layer, and on an area adjacent to the center of the wafer. Both differences may have caused the small discrepancy in thickness. We conclude from Table 2.2 Fig. 2.5 and Fig. 2.6 that thicknesses
20
measured by SE, HIM, HR-SEM and XRR are in good agreement. This confirms the validity of SE for measuring W films within their transparency range, i.e. up to 30 nm.
Importantly, the sub-nanometer thickness changes such as shown for example in Fig. 6.1 of Chapter 6 fall beyond the measurement accuracy of SE. These changes are in other words hardly physical and are only shown to indicate no measurable variation of the thickness during the corresponding experiments. The larger but still few-nm thickness changes (see Figures 6.2 and 6.3) solely indicate a qualitative trend (i.e., increase, decrease or hardly any change) in thickness behavior and do not provide quantitative information. The sub-monolayer numbers given in Table 6.1 can at best be interpreted as the average thickness over the mm-scale area probed by SE; this area features discrete nm-scale film islands on an otherwise uncovered surface.
A properly-tuned SE model still allows for a reliable monitoring of individual ALD or etching cycles, as demonstrated for example in Fig. 2.11. However, the within-cycle variations seen in this and similar figures could reflect the changes of optical properties of the surface upon exposure to different reactants and would not necessarily correspond to the actual (quantitative) thickness variations. To note, the net thickness increase (or decrease) after each cycle should still correspond to the actual film growth, thereby indicating the growth rate per cycle, provided the growth curve has been verified by other (ex-situ) techniques for each new material under study.
2.4. Delivery of at-H
Although at-H has to make a 90-degree turn in order to reach the substrate in both reactors, there is still an appreciable flux of at-H to the wafer surface in the
hot-wall reactor, as earlier confirmed by tellurium (Te) etching experiments[1]. Te
reacts with at-H at room temperature to form volatile hydrogen telluride (H2Te)[15];
21
Therefore, the observation of etching of a Te film in hydrogen ambient at room
temperature can confirm the existence of at-H at Te surface[1].
To verify the existence of at-H and its efficient delivery over 70 cm from the source to the substrate in the cold-wall reactor, Te etching experiments were carried out before starting the work on W film deposition in this reactor. The optical model
for a layer of Te on Si has been established and verified in our previous work[1] where
kinetics of Te etching was monitored by SE in real-time.
2.4.1. Te etching by at-H pulses in the cold-wall reactor
Our previous work, using the same hot-wire source installed in the hot-wall reactor, has demonstrated the high etch rate of Te by at-H, greatly depending on the
hot-wire temperature[1]. To enable HWALD, well-defined pulses of at-H must be
provided instead of a continuous flow. We therefore confirmed the ability to reliably supply at-H in pulses for HWALD in the following experiments (see Fig. 2.7). In
regime (1), the Te film could not be etched with the hot wire off while exposed to H2
flow (as expected). In regime (3), with hot wire on while exposed to argon or nitrogen, the Te film could not be etched either. The three regimes indicated in Fig. 2.7
manifest that only the combination of a switched-on hot wire with H2 flow along the
hot wire source causes etching of the Te film. It can be seen that Te etching occurs by each pulse of at-H of 0.1 s; the etching quickly diminishes and stops during a purge of 30 s with Ar. As the figure further indicates, there is no delay between the
injection of H2 and the on-set of etching; however etching tends to continue for
another 8 s after the 0.1 s pulse of hydrogen. This is attributed to the residence time of at-H in the reaction chamber. In conclusion, etching of Te films confirms not only the efficient delivery of at-H to the substrate but also its sufficiently long lifetime in the reactor.
22
Figure. 2.7. Etching a Te film by sequential 0.1 s pulses of at-H followed by 30 s purge. Shown is the
thickness of remaining Te on a silicon wafer as measured by SE. Conditions: room Ts, process pressure
of 0.003 mbar, H2 flow rate of 100 sccm and a carrier gas (Ar) flow rate of 50 sccm. Three regimes are
identified: (1) introducing molecular hydrogen (H2), hot wire off; (2) introducing H2 via the hot wire kept at
1750 °C; (3) Ar flowing through the hot wire at 1750 °C.
In this tellurium etching experiment, each 0.1s pulse of hydrogen could
etch approximately 0.3 nm of Te. Before calculating the dissociation efficiency of H2,
following assumptions are made: (i) all H2 molecules will dissociate on the hot wire,
(ii) all generated at-H could reach the surface to etch Te, (iii) no re-deposition of Te
occurred and (iv) etching was uniform and conformal across the wafer. With a H2
flow of 100 sccm, the rate of H2 molecules coming to the filament was (1 sccm gives
2.7 × 1018 H
2 molecules/min) 2.7× 1020 molecules/min. The pulse time was 0.1 s,
thus the number of molecules introduced in each cycle was around F=4.5 × 1017.
The number of Te atoms per nm of film thickness per cm2 is 1.5 × 1015 [16]. As two H
atoms are needed to form one H2Te molecule and thus etch one Te atom, 3.1 × 1015
H atoms were required to etch 1 nm of Te film per 1 cm2 of surface area. In each
pulse around 0.3 nm Te/cm2 was etched, requiring 9.8 × 1014 H atoms/cm2. As the
wafer area was around 81 cm2, the required number of hydrogen atoms to uniformly
etch the film was f= 7.9× 1016. Hence a dissociation efficiency of H
2 on the filament
surface can be roughly estimated at
23
A more accurate estimation requires chemical modeling of the etching reactions including knowledge of the sticking probability of at-H to the surface, the actual gas dynamics, reaction rate constants, and also taking the decomposition of the formed
TeH2 (i.e. re-deposition) into account. Unfortunately, several unknown parameters
of this process make it rather difficult to estimate the amount of at-H produced. In
terms of this simple model, 9.8 × 1014 H atoms/cm2 are needed (assuming 100%
participation in the reactions, sticking probability 1, etc.) to maintain the experimental etch rate. The actually generated amount of at-H could be higher (e.g., in case of a lower sticking probability) or lower (e.g., for a non-uniform etching with the etch front propagating gradually in the direction of flow).
2.4.2. Factors influencing on Te etch rate
The total process pressure is a crucial parameter affecting the etch rate. A higher process pressure indicates more species in the reactor and thus a higher chance for recombination of at-H in the gas phase. This combination is a three-body
process[17], in other words it requires three particles to collide at the same time. Any
Figure 2.8. Influence of the total process pressure on the removed thickness of Te per a-H pulse.
Conditions: room Ts, H2 flow rate of 100 sccm and a carrier gas (Ar) flow rate of 50 sccm. The hot-wire
24
surface or species inside the reactor can act as a third particle (body). As a result, the etch rate was severely suppressed with an increasing process pressure, confirmed in Fig. 2.8. Especially when the process pressure reached 1 mbar, etching was nearly terminated due to very limited at-H reaching the Te surface.
Flow rates of the carrier gas (Ar) can affect the partial pressure of at-H and thus the etch rate. Fig. 2.9 shows how the removed thickness of Te/pulse (a measure of the etch rate) reduced when a higher flow rate of Ar from the HW port was adopted. This was not only due to the dilution of at-H, but also the consequence of a higher probability of at-H recombination. Despite the negative effects on etch rate, a suitable dose of Ar was necessary to ensure an efficient purge. Similar effects of the Ar from the gas ring were expected.
Figure 2.9. Influence of Ar flow rate on the removed thickness of Te per a-H pulse. Conditions: room Ts,
process pressure of 0.003 mbar, H2 flow rate of 100 sccm and the hot-wire temperature was kept at
1750 °C.
2.4.3. Back-stream diffusion
In reality, except the flows of precursors coming from either hot-wire port or gas ring to the pump, there is also diffusion of precursors. It is possible that a precursor injected from the gas ring can diffuse upwards to the hot wire, a process
25
called back-stream diffusion here. To investigate the existence of back-stream
diffusion, H2 was introduced from the gas ring situated 10 cm above the substrate.
The occurrence of H2 diffusion upwards to the filament, the formation of at-H and its
subsequent transportation downwards to the substrate would result in Te etching. Fig. 2.10 confirms the existence of back-stream diffusion, by showing the etch rate as a function of the total process pressure and the at-H pulse time. Moreover, increasing the process pressure suppressed the etch rates. In comparison, at the
lowest pressure in Fig 2.10, the same H2 flow introduced from top of the reactor
resulted in an etch rate of 0.6 nm/pulse, whereas the etch rate by back-stream diffusion was 0.0375 nm/pulse (see the circled point). It indicates that, under the same experimental conditions, the contribution of back-stream diffusion to etching
was less than 10% compared to the forward-stream H2 flow. Minimizing the
back-stream diffusion is required for the optimization of HWALD experimental conditions; this will be considered later in this thesis.
Figure 2.10. Influences of total process pressure and at-H pulse time on the removed thickness of Te per
a-H pulse. Conditions: room Ts, H2 flow rate of 100 sccm, post-at-H purge of 1 min and the hot-wire
26
2.5. Hot wire is not a source for W deposition
To deposit W with assistance of a W filament, it is very important that the hot wire itself is not a source of tungsten. Since the W vapor pressure is rather low at
temperatures below 2000 °C[18], evaporation of W is negligible. A demonstration of
HWALD of W in the hot-wall reactor is presented in Fig. 2.11, where the hot wire was
kept at 1750 °C. In regimes (1) and (2), either Ar or Ar/H2 mixture was introduced via
the hot wire, showing no detectable growth of W. In regime (3), while exposed to
subsequent H2 and WF6 pulses, W started to grow in an ALD manner. The three
regimes indicated in Fig. 2.11 manifest that the W filament itself is not a source of W for deposition.
Figure. 2.11. A demonstration of W deposition by HWALD. Shown is the thickness of deposited W on W
surface measured by SE. Conditions: Ts of 325 °C, HW temperature of 1750 oC, process pressure of 0.05
mbar, H2 flow rate of 50 sccm and a carrier gas (Ar) flow rate of 50 sccm. Three regimes are identified:
(1) Ar flowing along the hot wire; (2) introducing H2 via the hot wire without introducing WF6; (3) introducing
27
2.6. Conclusions
This chapter describes the hot- and cold-wall reactors which will be used in the next experiments to grow HWALD W. The film thicknesses in-situ measured by spectroscopic ellipsometer have been verified by other ex-situ techniques. The optical models have been established and validated. This work further confirmed the existence of at-H and its delivery to the substrate in the cold-wall reactor under various experimental conditions. And here we also provide results of Te etching by at-H pulses in the cold-wall reactor. Despite a big volume and a long distance, at-H can transfer to the substrate surface and sustain a reasonable etch rate of Te. As demonstrated, the total process pressure, Ar flow rates and other parameters can affect etching and also influence the back-stream diffision of hydrogen.
References
[1] H. Van Bui, A. Y. Kovalgin, A.A.I. Aarnink and R.A.M. Wolters, J. Solid
State Sci. Technol. 2, P149 (2013).
[2] S. Bystrova, A. Aarnink, J. Holleman and R.A.M. Wolters, J. Electrochem.
Soc. 152, G522 (2005).
[3] P. Desai, T. Chu, H. M. James and C. Ho, J. Phys. Chem. Ref. Data 13,
1069 (1984).
[4] S. Heil, J. van Hemmen, C. Hodson, N. Singh, J. Klootwijk, F. Roozeboom,
M. van de Sanden and W. Kessels, J. Vac. Sci. Technol. A 25, 1357 (2007).
[5] S. Heil, J. van Hemmen, M. van de Sanden and W. Kessels, J. Appl. Phys.
103, 103302 (2008).
[6] E. Langereis, S. Heil, H. Knoops, W. Keuning, M. Van de Sanden and W.
Kessels, J. Phys. D: Appl. Phys. 42, 073001 (2009).
[7] H. Fujiwara, Spectroscopic ellipsometry: principles and applications (John
Wiley & Sons, New York, 2007).
[8] H. Tompkins and E. A. Irene, Handbook of ellipsometry (William Andrew,
New York, 2005).
[9] F. Wooten, Optical Properties of Solids (Academic Press, New York,
1972).
[10] P. W. Milloni and J. H. Eberly, Lasers (John Wiley and Sons, 1988).
28
[12] E. D. Palik, Handbook of Optical Constants of Solids (Academic Press,
1997).
[13] C. M. Hezinger, B. D. Jonhs, W. A. McGahan, J. A.Woollam and W.
Paulson, J. Appl. Phys. 83, 3323 (1998).
[14] G. Jellidon and F. Modine, Phys. Rev. B 27, 7466 (1983).
[15] D. A. Outka, Surf. Sci. 235, L311 (1990).
[16] M. K. Slattery, Physical Review 25, 333 (1925).
[17] H. Wise and C. M. Ablow, J. Chem. Phys. 35, 10 (1961).
[18] R. Szwarc, E. Plante and J. Diamond, NBS J. Res. Phys. Chem. 69, 417
29
3
Abstract.
In this chapter, tungsten films were deposited in the cold-wall reactor by HWALD. The real-time SE monitoring revealed the co-existence of three processes: CVD, etching and ALD of the W film. WF6 could back-stream diffuse to the hot-wire, resulting in WF6
decomposition and generation of a flux of fluorine (F). The latter caused etching of the grown W film and the filament, and provided extra tungsten supply, which might cause CVD. By controlling the dose of WF6 and process pressure, the etching had been minimized. Further,
we compared samples with tungsten grown by either HWALD or chemical vapor deposition (CVD) in terms of growth kinetics and properties. For CVD, the samples were made in a mixture of WF6 and molecular or atomic hydrogen. Resistivity of the CVD W was around 20
µΩ·cm, whereas it was as high as 100 µΩ·cm for the HWALD films. X-ray diffraction (XRD) reveals that the HWALD W was crystallized as β-W, whereas both CVD films were in the α-W phase.
This chapter is based on the publications:
Mengdi Yang, Antonious A.I. Aarnink, Alexey Y. Kovalgin, Rob A.M. Wolters and Jurriaan Schmitz, “Hot-wire assisted ALD of tungsten films: In-situ study of the interplay between CVD, etching and ALD modes”, Phys. Status Solidi A, 212, 1607 (2015).
Mengdi Yang, Antonious A.I. Aarnink, Alexey Y. Kovalgin, Dirk J. Gravesteijn, Rob A.M. Wolters and Jurriaan Schmitz, “A Comparison of Tungsten Films Grown by CVD and Hot-wire Assisted ALD in a cold-wall reactor”, J. Vac. Sci. Technol. A, 34, 01A129 (2016).
30
3.1. Introduction
In this chapter, results of tungsten formed by HWALD using sequential WF6
and H2 in the cold-wall reactor are presented. During optimization of the HWALD
window, an interplay among CVD, etching, and ALD modes was found and investigated. The interplay is highly dependent on deposition conditions. By optimizing these conditions, the HWALD mode can be successfully enabled. Film properties, such as resistivity, thickness, roughness, density, elemental composition and crystal structure have been evaluated by means of four-point probe (FPP), atomic force microscope (AFM), X-ray diffraction (XRD) and reflection (XRR), X-ray photoelectron spectroscopy (XPS), high resolution transmission electron microscopy (HR-TEM) and HR-SEM. Further, we compared three deposition methods of W: (i)
CVD by using WF6 and molecular hydrogen (H2-CVD), (ii) CVD by WF6 and atomic
hydrogen (at-H-CVD) and (iii) HWALD.
3.2. Experimental
3.2.1. Deposition
In this experiment, tungsten films were deposited on a 100-nm-thick silicon
oxide (SiO2) thermally grown on p-type Si (100) substrate. Prior to deposition, the
wafers were cleaned in fuming (99%) HNO3 and boiling 69% HNO3 to remove
organic and metallic contaminations, respectively. Finally, the substrate was
immersed into a 0.3% HF solution for 3 min.As the tungsten nucleation is very poor
on SiO2 and Si3N4[1], a seed layer is required[2]. Formation of this seed layer consisted
of (i) CVD of a thin (<10 nm) layer of amorphous Si (a-Si) from trisilane (Si3H8); and
(ii) converting the a-Si into W by reacting with WF6. The whole process of forming
the seed layer was performed in the hot-wall reactor, and then the wafer was transferred to the cold-wall reactor without vacuum break.
To be more specific, a Si3H8 flow of 10 sccm was supplied. The precursor
31
pressure of 10 mbar, forming a layer of amorphous silicon (a-Si) of approximately 5 nm. The thickness of a-Si was monitored in-situ by SE. Next, the reactor was
pumped down to a pressure of 1 mbar followed by introducing WF6 (10 sccm) to
convert the a-Si into W. Further, the reactor was flushed by N2 for 10 min and
evacuated back to 10-7 mbar, followed by the wafer transfer to the cold-wall reactor,
to continue with the HWALD process.
The HWALD of W films was carried out using sequential pulses of WF6 and
at-H at substrate temperatures ranging between 200 and 325 °C, and process pressures of 0.001 –1 mbar. The hot-wire temperature was varied between 1800 to 1900 °C. For ex-situ (electrical) characterization, it is important to prevent oxidation of the grown W film during its exposure to air. The wafer was therefore transferred back to the hot-wall reactor after the HWALD process, where a 10 nm capping layer of a-Si was deposited (employing the same deposition method as used for the a-Si seed layer) on top of W, completing the process.
3.2.2. Film characterization
The film thickness was measured in real-time during the deposition using the in-situ spectroscopic ellipsometer described in Chapter 2. The measurements were taken every 2.5 s. Due to the opacity of thicker W layers, only the films with thickness up to 30 nm could be measured by SE. The film thickness was verified by high-resolution scanning electron microscopy (HR-SEM) and XRR, shown earlier in Chapter 2. Additionally, optical properties of the films were also obtained by SE.
The elemental composition of the as-grown film was obtained by XPS. A PANalytical X’PERT MPD diffractometer was utilized for XRD and XRR measurements. XRD and XRR patterns of the samples were recorded in the region of 2θ = 30-90° using Cu Kα radiation, with a PANalytical PIXcel1D detector. The film resistivity was examined by an automatic four-point probe stage of Polytec. The film surface morphology was characterized by a Bruker Fastscan/ICON model AFM.
32
Further, HR-TEM and energy filtered TEM (EFTEM) were utilized to characterize grown films.
3.3. In-situ study of the interplay between CVD, etching
and ALD modes
3.3.1. The existence of CVD and etching modes
According to literature, the thermal dissociation of WF6 (in gas phase) starts
at 750 °C[3,4]. Although the gas ring was installed at a distance of 70 cm below the
hot wire, the operating conditions (pressure, gas flows) could greatly influence the
upward diffusion of WF6 to the hot wire. As a consequence, this WF6 may dissociate
at the hot wire, likely generating tungsten subfluorides (WFx) and fluorine-containing
species (atomic fluorine and fluorine)[5-8]. As the dissociation of WF6 occurs in
equilibrium, any increase of concentration of WF6 will enhance the generation of
fluorine-containing species. It is reported that the lifetime of atomic fluorine in a
(plasma) reactor can be quite long[9]. Additionally, both WF
x and fluorine-containing
species can adsorb on the (cold) reactor walls[10,11]. This provides a background
supply of WF6 and fluorine-containing species to the gas phase over the deposition
cycles. Fluorine-containing species, generated from WF6 dissociation upon the hot
wire, can contribute to etching the hot wire and regeneration of volatile tungsten precursor(s) (likely tungsten fluorides). When mixed with at-H, these precursors may contribute to a CVD mode at substrate level. The experiments shown in Fig. 3.1
confirm the existence of a CVD mode. For example a 1-min exposure to WF6 gas
followed by a 2-min purge and subsequent at-H exposure reveals a continuous growth of W film up to 2.5 nm in thickness (filament temperature of 1910 °C). This occurs during the entire (up to 25 min) exposure to at-H only, likely due to CVD.
33
(a) (b)
(c) (d)
Figure 3.1. Film growth (obtained by in-situ SE) versus different deposition parameters. All long-time
exposures to at-H were performed after a 1-min exposure to WF6 gas followed by a 2-min purge. Standard
parameter values: 0.01 mbar of pressure, 325 °C of Ts, 1860 °C of hot-wire temperature, 100 sccm of H2,
10 sccm of WF6, 100 sccm of H2-carrier gas (Ar), and 50 sccm WF6-carrier gas (Ar). Each graph shows
the influence of one parameter only while keeping the standard values for all other parameters. The Figure shows the influence of: (a) hot-wire temperature, (b) H2 flow rate, (c) WF6 flow rate, and (d) H2-carrier-gas
(Ar) flow rate.
Figures 3.1 (a-d) show the CVD growth under different conditions, while Ts
and total pressure were fixed at 325 °C and 0.01 mbar. Firstly, in Fig. 3.1 (a) the growth rate increases with hot-wire temperature due to the improved dissociation
efficiency of H2[12], resulting in a higher at-H flux. However, the difference between
1860 °C and 1910 °C is small, probably due to the limited reactant supply (i.e. volatile tungsten precursors). From Fig. 3.1 (b) it can be seen that there is no
significant impact of the H2 gas flow rate on CVD growth rate, likely pointing to a
comparable amount of at-H reaching the substrate surface in these cases. However,
our experiments (not shown) indicate that the upstream diffusion of WF6 can be
34
minimize CVD, the flow rate of H2 is determined to be 100 sccm (the maximum flow
rate in our case). From Fig. 3.1 (c), a higher flow of the initial WF6 enhances the
growth rate. It is noticeable that the growth rate in the first 5 min doubles when the
WF6 flow rate becomes two times larger, implying that the initial 0-to-10-min growth
is mainly influenced by the WF6 injected from the gas ring. On the contrary, growth
rates for times exceeding approximately 20 min are almost the same despite the
different WF6 flow rates. Analyzing the curves (values and shapes), one can
conclude that the growth rates after an exposure of around 20 min follow the kinetics of desorption of the reactants that adsorbed on the cold walls during the preceding
WF6 pulse(s). In Fig. 3.1 (d), it is shown that the flow rate of H2-carrier gas (Ar) has
a significant effect on the CVD growth rate. Apparently, a higher downward flow rate could (i) suppress the upward diffusion of the reactants, thereby decreasing their interaction with the hot wire, (ii) decrease the partial pressures of reactants and thus suppress the related CVD mode, and (iii) shorten the delivery time of at-H to the substrate. As a matter of fact, a 10-times higher Ar flow decreases the growth rate approximately 2 to 4 times, depending on the time of exposure to at-H.
To examine the film grown by the CVD process shown in Fig. 3.1, X-ray photoelectron spectroscopy (XPS) was used to obtain the elemental composition of the selected W film. This film was grown at 0.01 mbar with the following cycle
sequence: 2.5 min WF6/ 2 min purge/ 10 min at-H/ 2 min purge. Flow rates of WF6
and H2 were 5 and 100 sccm, respectively. The hot-wire temperature was 1750 °C,
and the substrate temperature was 325 °C, Ar-carrier-gas flow rates for WF6 and H2
were 50 and 100 sccm, respectively. One should note the remarkably low concentrations (i.e. at detection limits) of all impurities and the concentration of tungsten at 92-98 at%. It is quite clear that the continuous deposition of W films up to 2.5 nm is evidence of a permanent source of tungsten in the gas phase, most
probably in the form of WF6. As mentioned above, this source is expected due to
35 Figure 3.2. XPS-profile of the film grown at 0.01 mbar with the following cycle sequence: 2.5 min WF6/
2 min purge/ 10 min at-H/ 2 min purge. Flow rates of WF6 and H2 were 5 and 100 sccm, respectively.
Other conditions: hot-wire temperature of 1750 °C, Ts of 325 °C, Ar-carrier-gas flows for WF6 and H2 of
50 and 100 sccm, respectively. The Si signal at the beginning is from the a-Si capping layer.
Figure 3.3. CVD growth rate versus process pressure during a 1 min exposure to at-H, after a 1-min
exposure to WF6 gas followed by a 2-min purge. Conditions: Ts of 325 °C, hot-wire temperature of 1750 °C,
H2 of 100 sccm, WF6 of 2 sccm, H2-carrier gas (Ar) of 100 sccm, and WF6-carrier gas of 50 sccm (Ar).
The impact of the total gas pressure on the growth rate was additionally studied, see Fig. 3.3. On one hand, a higher pressure is expected to increase the
recombination of at-H[13] and to suppress the upward diffusion. On the other hand, a
higher (partial) pressure of WF6 will shift the equilibrium towards generating more
36
expected. This is confirmed in Fig. 3.3: a pressure of 0.3 mbar sufficiently decreases the CVD growth.
To conclude, the H2 and top-carrier-gas flow rates were chosen both at
100 sccm (maximum values), in combination with high enough process pressure, to
minimize the possible CVD mode. Moreover, the WF6 pulse should be kept short, to
limit the unwanted supply of reactants enhancing CVD. This can be achieved by
limiting WF6 pulse time and flow rate.
3.3.2. Etching mode of deposited W films
Fluorine, generated by the decomposition of WF6 on the hot wire, will diffuse
downwards and etch the deposited W film. This effect was more pronounced at high
WF6 pressures, due to the mentioned shift of equilibrium. The interplay between
growth and etching modes is presented in Fig. 3.4. The high sensitivity of in-situ SE
allows study of the separate ALD cycles (0.1 s at-H, 60 s purge, 0.1 s WF6, and 60 s
Figure 3.4. In-situ monitoring of individual pulses by SE. Black lines indicate the slopes. The black arrows
show the admittance of the corresponding gas pulses to the reactor. One cycle consisted of 0.1 s at-H exposure, 60 s purge, 0.1 s WF6 exposure, and 60 s purge. Conditions: pressure of 1 mbar, Ts of 325 °C,
hot-wire temperature of 1750 °C, 100 sccm Ar carrier gas of H2, 50 sccm Ar carrier gas of WF6, and
100 sccm H2 flow and 10 sccm WF6 flow.. Due to the short pulse times of at-H and WF6, the last two flow
37
purge). Pulses of at-H and WF6 are indicated. One can clearly see a gradually
increasing decay of the thickness after introducing every next WF6 pulse, i.e. etching.
In the first two cycles starting from 0.23 nm at 7 minutes, the growth rate (H2 pulse)
is approximately equal to the etch rate (WF6 pulse), resulting in near-zero net growth
of the film. From the third cycle, etching starts to increasingly dominate, leading to the net decline of the film thickness. This behavior, typical and reproducible for given conditions, can be explained by a gradual increase of fluorine concentration in the
reactor due to the preceding WF6 pulses. This highlights the importance of limiting
the WF6 dose during each pulse, in order to minimize etching.
The influence of substrate temperature (Ts) and total gas pressure on the
net growth is shown in Fig. 3.5, which further illustrates the interplay between deposition and etching. It is obvious that etching is enhanced by a higher pressure resulting in a lower growth rate. Furthermore, etching becomes dominant at a lower temperature if the pressure is higher. Considering etching and deposition as two
parallel reactions, an increase in Ts results in a higher etch rate rather than a higher
growth rate, we conclude that etching has a stronger dependence on Ts compared
to deposition. In order to achieve a higher growth rate, 315 °C was determined to be
the Ts for the following ALD experiments.
Figure 3.5. Influence of total pressure and substrate temperature on net growth. The cycle sequence:
10s H2 (100 sccm)/ 30 s Ar (purge)/ 0.1 s WF6 (2 sccm)/ 30 s Ar (purge). Other conditions: Ar-carrier gas
flow rate of 100 sccm and 50 sccm for H2 and WF6, respectively; hot-wire temperature of 1750 °C. Due to